Programmable Surfaces and IRS Technology

• Programmable surfaces are engineered metasurfaces composed of tunable meta-atoms that modify electromagnetic waves for customizable propagation environments.
• They enable advanced wireless and sensing applications via techniques like electronic, optical, phase-change, and MEMS tuning with performance gains up to 15 dB and enhanced SNR.
• IRS technology integrates robust signal processing and optimization methods to reduce pilot overhead, improve spectral efficiency, and support scalable, multi-hop deployments.

A programmable surface, broadly referred to as an Intelligent Reflecting Surface (IRS) or Reconfigurable Intelligent Surface (RIS), is an engineered, planar structure comprising a two-dimensional array of passive or semi-passive elements, each capable of imparting programmable phase and/or amplitude modifications to impinging electromagnetic fields. By jointly coordinating these programmable elements, IRSs enable deterministic, software-defined manipulation of the propagation environment for wireless communications and sensing applications. This paradigm, building upon advances in metasurface engineering, digital control, and network optimization, is considered foundational for future wireless architectures extending into the terahertz (THz) regime, massive MIMO, and integrated sensing-communication systems.

1. Physical Principles, Architectures, and Hardware Implementations

Programmable surfaces operate by spatially and temporally modulating incident electromagnetic waves through arrays of sub-wavelength elements—meta-atoms—whose electromagnetic boundary conditions are tunable by external stimuli. Four principal THz-band IRS tuning mechanisms have been rigorously established (Wu et al., 20 Jun 2025):

1. Electronic (CMOS/Schottky/Graphene): Programmability derives from tuning carrier concentrations via voltage bias (e.g., in Schottky-junction split-ring resonators, CMOS-inverter-driven phase shifters, or graphene sheets). CMOS-based IRSs (e.g., 65 nm node) achieve up to 8-bit granularity for amplitude/phase, nanosecond reconfiguration, amplitude range up to 25 dB, and phase range ±30° at 0.3 THz.
2. Optical (Photoconductive): Illumination by sub-bandgap optical pulses photo-generates carriers in semiconductors (e.g., high-resistivity Si or GaAs meta-atoms), shifting their complex refractive indices on picosecond scales. Ultrafast, but energy costs and thermal loads are pronounced.
3. Phase-Change Materials (PCMs, e.g., VO₂, GeSbTe, Liquid Crystals): Local heating or optical/electrical pulses induce phase transitions, producing nonvolatile, multi-level analogue index control; e.g., VO₂ transitions at ~68°C, enabling persistent logic states and analog phase control.
4. MEMS (Micro-Electro-Mechanical Systems): Mechanical deformation of meta-structures (such as cantilevers) actuated electrostatically, enabling large phase ranges, low loss, but generally limited by microsecond-to-millisecond reconfiguration times.

For reflectarray- or patch-based IRSs, general architecture consists of multi-layer PCBs combining RF and DC biasing layers, with each unit cell (or “patch”) incorporating tunable reactances (e.g., varactor diodes, PIN diodes) for phase control. Modularity, as demonstrated in distributed microcontroller-based 1-bit RIS arrays, enables aperture scaling without complex wiring or circuitry, leveraging wireless (e.g., infrared) digital control (Sayanskiy et al., 2022).

2. Reflection and Channel Models

Each IRS element $m$ is modeled as a zero-thickness scatterer defined by its local reflection coefficient:

$$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$

with $\alpha_m \in [0,1]$ the amplitude response and $\phi_m \in [0,2\pi)$ the programmable phase shift (Wu et al., 20 Jun 2025).

Far-field phase law:

$$\phi_m(\theta_0, f_0) = -\frac{2\pi}{\lambda_0}(x_m \sin\theta_0 + y_m \cos\theta_0) + \phi_m^{\rm prog}$$

where $(x_m,y_m)$ is the meta-atom position, $\lambda_0 = c/f_0$ the wavelength, and $\phi_m^{\rm prog}$ the programmed state. Under wideband excitation, “beam squint” arises due to frequency-independent phase control: actual beam directions drift with frequency as the ideal steering law is frequency-scaling, but realized shifts are not.

Cascaded channel (BS–IRS–user): Using standard notation, the composite effective channel under IRS is modeled as:

$$\mathbf{h}_{\rm eq} = \mathbf{h}_d + \mathbf{h}_r^H \boldsymbol{\Phi} \mathbf{H}_t$$

where $\mathbf{H}_t$ is the BS–IRS channel, $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$0 the IRS–UE channel, and $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$1 the IRS phase diagonal (Pan et al., 2020, Wu et al., 15 Jan 2025). Extended models exist for double/multi-hop IRS chains and distributed IRS deployments (Wu et al., 15 Jan 2025, Yashvanth et al., 2024).

3. Near-Field, Wideband, and Spatial Effects

Near-field regime: The Fresnel (near-field) boundary for an IRS of aperture diameter $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$2 and wavelength $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$3 is $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$4 (Wu et al., 20 Jun 2025, Hibi et al., 2024). For typical THz IRSs ($$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$5 m, $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$6 mm), Fresnel boundary $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$7 m—user links at tens of meters are typically near-field.

Beamfocusing vs. Beamforming: In the near-field ($$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$8), optimal IRS phase patterns require spherical wavefront focusing (dependence on both direction and user range), not standard far-field plane-wave steering. Phase errors from far-field approximations in the near-field can produce significant main-lobe degradation, beam pattern distortion, and reduced link gains (Hibi et al., 2024).

Double beam squint: In wideband cascaded (BS–IRS–UE) settings, both the BS and IRS induce frequency-dependent pointing shifts (“beam squint”), resulting in frequency-dependent array gains and directionality that complicate both beam alignment and channel estimation (Wu et al., 20 Jun 2025).

4. Signal Processing, Codebook, and Optimization Approaches

Hierarchical beam training: IRS-assisted links exploit coarse-to-fine, two-stage beam codebooks: coarse angular search followed by joint angular/distance (“3D”) refinement (codebook size scaling with $$r_m(\theta, f) = \alpha_m(\theta, f) e^{j\phi_m(\theta, f)},$$9). Nonuniform, cluster-optimized 3D codebooks, constructed via k-means on a control accuracy index, offer SNR gains and overhead savings compared to naive (uniform or 2D) codebooks (Hibi et al., 2024).

Codebook-based IRS control: For large IRSs, phase-shift codebooks (with $\alpha_m \in [0,1]$0 codewords for $\alpha_m \in [0,1]$1 elements) are designed to maximize the worst-case beamforming response within angular “tiles.” Both continuous (optimized via rank-penalized SDPs) and discrete ($\alpha_m \in [0,1]$2-bit phase) codebooks can be constructed with polynomial- or MILP-based approaches; 2-bit quantization suffices to closely match continuous designs, greatly reducing online control overhead (Ghanem et al., 2022).

End-to-end optimization: With programmable IRSs, typical objectives include maximizing sum-rate, minimizing transmit power for prescribed SNR targets, or optimizing minimum-secrecy-rate under eavesdropper constraints. Due to the unit-modulus phase constraint and multiplicative coupling with transmit beamformers, these are nonconvex problems, commonly solved using alternating optimization (AO), semidefinite relaxation (SDR), manifold optimization, or branch-and-bound schemes for modest dimensions (Yu et al., 2020, Wu et al., 15 Jan 2025, Chen et al., 2019).

Robust optimization: In presence of channel uncertainty and imperfect CSI, robust IRS designs account for mean-square error, pilot overhead, and statistical mismatch, commonly using Bayesian MMSE or compressed sensing-based estimation. CSI acquisition overhead can be reduced by exploiting spatial/angular sparsity or by jointly optimizing IRS phase training sequences (Alwazani et al., 2020, Rahim et al., 2024).

5. System-Level Design and Deployment Strategies

Deployment scales and architectures:

• Large-scale (macro-IRS): Static, high-element-count surfaces deployed to intercept and redirect unfavorable propagation, typically near BS or hotspot user clusters; distributed patches can increase spatial DoFs and reduce beam squint (Wu et al., 20 Jun 2025, Wu et al., 15 Jan 2025).
• Small-scale, movable IRSs: Modular panels on mobile robots, UAVs, or movable platforms enable agile 3D beam steering, blockage avoidance, and dynamic adaptation—a key for IIoT and mmWave/THz deployments (Gao et al., 14 Jan 2026).
• Multi-hop/Cooperative IRSs: Chained or distributed IRSs, possibly in cooperation or relay-aided architectures, can achieve higher-order beamforming gains ($\alpha_m \in [0,1]$3 for $\alpha_m \in [0,1]$4-hop chains), rank augmentation, and coverage extension with fewer elements by trading array gain and active power (Ying et al., 2020, Wu et al., 15 Jan 2025).

Practical considerations:

• Hardware constraints: Finite phase resolution, insertion loss, slow reconfiguration, and control line complexity impact system performance. Practical prototype IRSs in the 220 GHz regime demonstrate ≥15 dB power gains and sub-beam EVM improvements of 40–60% in multi-user OFDM systems (Wu et al., 20 Jun 2025).
• Control architecture: Distributed, block-level control with addressable microcontrollers (as in IR-controlled RIS modules for Wi-Fi) provides scalable, hot-swappable apertures with robust synchronization and remote reconfiguration (Sayanskiy et al., 2022).
• Cost and deployment trade-offs: Optimization frameworks employing moving antennas and IRS site selection minimize capital and operational expenditure, balancing SNR coverage against hardware, installation, and RF-chain costs (Gao et al., 14 Jan 2026).

Distributed IRS and network effects: In multi-operator environments, distributed IRSs benefit all operators by providing logarithmic scaling of spectral efficiency with total element count, exponentially reducing outage as the number of IRSs grows, and delivering spill-over gains irrespective of scheduling (Yashvanth et al., 2024).

6. Applications, Performance Benchmarks, and Prototyping Results

Performance enhancements: Programmable IRSs routinely yield:

• SNR and spectral efficiency improvements up to tens of dB (with $\alpha_m \in [0,1]$5 elements): e.g., downlink $\alpha_m \in [0,1]$6 up by 15 dB, uplink SNR up by 18 dB, and two-fold increases in spectral efficiency in 6G IoT setups (Mahbub et al., 2022).
• Robust minimum-secrecy-rate maximization, even under quantized or discrete phase constraints, using AO/path-following algorithms with convergence guarantees; 3-bit phase quantization is sufficient for high secrecy performance (Chen et al., 2019, Zhou et al., 2024).
• Substantial reduction in required element count and pilot overhead for relay-aided IRS structures and distributed channel estimation designs (Ying et al., 2020, Alwazani et al., 2020).

Experimental validations: Prototypes at 220 GHz with 80-element liquid-crystal IRSs and programmable phase panels demonstrate received power gains >15 dB and BER reductions from $\alpha_m \in [0,1]$7 to $\alpha_m \in [0,1]$8 at 10 dB SNR, substantiating the ability to extend links and support dense multi-user modulation formats (Wu et al., 20 Jun 2025).

7. Open Challenges and Research Directions

Critical open areas include:

• Near-field and wideband IRSs: Developing spherical wave-based channel models, nonuniform codebooks, low-memory hierarchical beamtraining, and beam squint compensation for ultra-large arrays and wideband signals (Hibi et al., 2024, Wu et al., 20 Jun 2025).
• Scalable channel estimation: Minimizing pilot/CSI overhead for massive passive arrays by exploiting sparsity or learning-based channel inference (Wu et al., 20 Jun 2025, Alwazani et al., 2020, Rahim et al., 2024).
• Multi-functional, cooperative, and secure IRSs: ISAC (Integrated Sensing and Communication), secure beam design for PLS, and coalition-formation games involving third-party IRSs require cross-layer and game-theoretic optimization (Wu et al., 14 Nov 2025, Zhou et al., 2024).
• AI-driven, adaptive control: Real-time IRS adaptation for user mobility, environmental changes, or multi-metric objectives (e.g., power/sensing/tracking) leveraging deep reinforcement learning and unsupervised control approaches (Wu et al., 14 Nov 2025).
• Quantum and optical extensions: Applications in FSO (Free-Space Optical) links—where IRSs function as smart mirrors—pose new challenges regarding misalignment robustness, mechanical/thermal jitter, and geometric optimization (Najafi et al., 2019).

System-level integration, hardware-software co-design, and experimental demonstration remain essential for the realization and widespread adoption of large-scale programmable surfaces in next-generation wireless and sensing ecosystems.